AI and Auditors: What's Actually Happening and What to Do

The honest assessment

If you work in audit, you've been hearing about how technology will change your job for at least a decade. Data analytics. Continuous auditing. Process mining. Most of it never quite delivered on the promises. AI is different. This time the technology actually works.

Here's what AI can do in audit right now. Analyse entire populations of transactions rather than samples. Identify anomalies, outliers, and patterns in financial data that humans miss. Review contracts and agreements for compliance with accounting standards. Draft audit reports and management letters. Generate risk assessments based on industry data and financial analysis. Summarise complex regulatory guidance. Process and categorise evidence. KPMG's AI tools can analyse millions of journal entries and flag unusual patterns. Deloitte has deployed AI across its audit practice globally. EY's AI-powered tools are performing audit procedures that used to take teams of juniors weeks.

What AI can't do is exercise professional scepticism in the way a seasoned auditor does. It can tell you that a number looks unusual. It can't sit across the table from a finance director who's explaining the unusual number and sense that they're lying. It can't read the dynamic between the CEO and CFO and recognise that the CFO is uncomfortable with the revenue recognition approach but won't say so in front of the CEO. It can't make the judgement call about whether a going concern issue is real or manageable based on knowledge of the client's industry, their competitors, and the economic environment. Professional judgement in audit is an accumulation of experience that doesn't reduce to data points.

The trajectory is clear though. The volume work of audit... testing samples, ticking and bashing, vouching transactions, reviewing calculations... is being automated systematically. The Big Four are investing billions because they've done the sums. Literally. The question isn't whether AI will transform audit. It's how many auditors the transformed version needs.

Your exposure level: High

High exposure. Audit work is structured, rules-based, and data-heavy. It involves comparing what is against what should be, and flagging the gaps. That's an AI task description.

The numbers support this. PwC committed over $1 billion to its AI strategy, with audit being a primary focus. KPMG, Deloitte, and EY have made similar investments. They're not spending that money to make their existing audit teams slightly more efficient. They're spending it to fundamentally change how many people are needed to complete an audit engagement.

The most affected roles are at the bottom of the audit hierarchy. Graduate trainees and audit seniors who spend most of their time on fieldwork... testing transactions, reviewing calculations, preparing workpapers... are directly in the firing line. The work they do is precisely the work AI automates. Some firms are already hiring fewer audit trainees. Not because there's less work, but because AI handles a larger proportion of the fieldwork.

For audit managers and partners, the exposure is lower but still significant. AI changes what you need to manage (more technology, fewer people), what you need to review (AI output rather than human workpapers), and what clients expect (faster, cheaper, more insightful audits). The role doesn't disappear but it transforms, and not everyone will make the transition successfully.

There's a silver lining buried in here. Audit has been widely considered boring and exhausting. The parts AI is automating are the parts most auditors hate: the repetitive testing, the endless sample selection, the documentation treadmill. What remains is the interesting stuff: risk assessment, professional judgement, client relationships, and the genuinely challenging accounting issues. If the profession plays this right, the job gets better. It just employs fewer people.

The 90-day action plan

1. This week: test AI on an audit procedure. Take a recent workpaper. Feed the underlying data into Copilot or ChatGPT and ask it to perform the analysis. Compare the output to your workpaper. For straightforward testing (recalculation, comparison, completeness checks), the AI will probably match or exceed your work. For judgement-heavy procedures, it won't. Map that boundary.

2. Week two: use AI for risk assessment. Give Claude the financial statements and background information for a client you know well. Ask it to identify the key audit risks and explain its reasoning. Compare to your firm's risk assessment. It'll identify the obvious risks well and occasionally flag something you hadn't considered. It'll also miss risks that require insider knowledge.

3. By day 30: automate your reporting. Use ChatGPT to draft management letters and audit reports from your findings. Feed it the issues, the implications, and the recommendations. The first draft quality will be 75-80% there. Your value is in the final 20-25%: the tone, the prioritisation, the specific recommendations that show you understand the client's business.

4. By day 45: learn full-population testing. If your firm has data analytics tools, learn them properly. If it doesn't, propose adopting one. The shift from sample-based to full-population audit is the single biggest change in audit methodology in a generation. Auditors who understand how to design and execute full-population tests using AI are ahead of the curve.

5. By day 60: deepen your industry expertise. Pick a sector and go deep. Technology. Financial services. Healthcare. Manufacturing. The auditor who understands the specific risks, accounting issues, and regulatory environment of an industry can do things AI can't: apply contextual knowledge to audit judgements. Generalist auditors are more replaceable than specialists.

6. By day 75: build your client relationships. AI can audit the numbers. It can't have the awkward conversation with the FD about their impairment assumptions. It can't navigate the politics of an audit committee meeting. It can't build the trust that means a client calls you when they're uncertain about an accounting treatment before they commit to it. Invest in the human side.

7. By day 90: get visible in your firm. Present on AI in audit. Write an internal paper. Volunteer for the firm's innovation committee. The people who influence how AI is implemented in audit practice have more career security than those who simply use whatever tools are given to them.

AI tools you should be using this week

• Microsoft Copilot for Work — Essential for Excel-heavy audit work. Data analysis, ratio calculations, trend analysis, and workpaper preparation are all faster with Copilot. Also useful for summarising client correspondence in Outlook and drafting reports in Word.

• ChatGPT for Work — Use it for drafting management letters, audit reports, and client communications. Also good for quick research on accounting standards and generating audit programme steps for unfamiliar areas. The ability to explain complex accounting concepts in plain English is useful for client-facing situations.

• Claude for Work — Better for long document analysis. Paste in a full set of financial statements, a complex lease agreement, or a regulatory filing and ask specific audit-relevant questions. Handles technical accounting discussions well and is good at identifying inconsistencies between different documents.

• Perplexity for Research — Current information on industry trends, regulatory changes, and market conditions. Essential for updating your risk assessment with real-time information. When a client's industry is going through a significant change, Perplexity briefs you faster than any other tool.

What to say in meetings

In partner meetings: "I've been using AI tools to reduce the time spent on routine testing by about 40%. The quality of our risk analysis is actually up because I'm spending more time on the judgement areas and less time on the mechanical work. Happy to share the workflow."

When clients ask about AI in audit: "We use AI to analyse your data more comprehensively than sampling ever could. That means better coverage and fewer surprises. The professional judgement, the risk assessment, and the reporting... that's still us." Clients want to know their audit is thorough AND that qualified people are making the decisions.

At audit committee meetings: "Our AI-powered analysis identified [specific finding] that traditional sampling might have missed. This is an example of how technology is improving audit quality, not just efficiency." Give them a concrete win from AI to build confidence in the approach.

If the worst happens

If you're made redundant from an audit role, your skills transfer to a surprising number of areas. Financial analysis, risk management, compliance, internal audit, consulting, and corporate governance all draw on the same core capabilities. You understand financial statements, regulatory requirements, and how businesses actually operate under the surface. That's valuable everywhere.

Natural adjacent moves: internal audit (companies always need it), risk management, financial controller, compliance officer, forensic accounting, or consulting. The Big Four's own consulting arms often recruit from their audit practices because auditors understand businesses from the inside out. Regulatory bodies (FCA, PRA, FRC) also recruit experienced auditors for their supervisory functions.

The honest observation is this. Audit firms have always had high turnover. The pyramid model was designed for it. But historically, people left audit because they wanted to, usually for better-paid roles in industry. The shift is that some people will leave because there isn't a role for them anymore. That's a different experience, and it hurts differently. If it happens to you, remember that the skills you built in audit... analytical rigour, professional scepticism, the ability to ask difficult questions and handle difficult answers... are not common skills. They're in demand. They just might not be called "auditing" in your next job. That's fine. The skill is the same. The label doesn't matter.